The traditional safety distance model based on the braking process is inadequate for scenarios where the adhesion coefficient varies. This research presents a safety distance model that takes into account the real-time fluctuations in the adhesion coefficient during the braking process. The proposed safety distance model is the basis for a three-level control strategy aimed at ensuring the safety and stability of the car-following process. A simulation is conducted to model the variation of the road’s adhesion coefficient during car braking. The experimental results, obtained for various initial adhesion coefficients and speeds, demonstrate that the precision of the safety distance calculated is improved by a minimum of 14.99% when compared to the traditional safety distance. Furthermore, the simulation results of the car-following process of a motorcade consisting of five cars provide evidence of the safety and stability of the suggested model.

Risk prediction is an important task to ensuring the driving safety of railway trams. Although data-driven intelligent methods are proved to be effective for driving risk prediction, accuracy is still a top concern for the challenges of data quality which mainly represent as the unbalanced datasets. This study focuses on applying feature extraction and data augmentation methods to achieve effective risk prediction for railway trams, and proposes an approach based on a self-adaptive K-means clustering algorithm and the least squares deep convolution generative adversarial network (LS-DCGAN). The data preprocessing methods are proposed, which include the K-means algorithm to cluster the locations of trams and the extreme gradient boosting recursive feature elimination based feature selection algorithm to retain the key features. The LS-DCGAN model is designed for sparse sample expansion, aiming to address the sample category distribution imbalance problem. The experiments implemented with the public and real datasets show that the proposed approach can reach a high accuracy of 90.69%, which can greatly enhances the tram driving safety.

目的异质遥感图像由不同传感器获取，在成像机理、辐射响应和空间纹理等方面存在显著差异，为获取变化区域信息而进行直接比对易产生大量“伪变化”。同时，传统无监督跨模态学习未区分像素状态，真实变化像素参与全局对齐训练会干扰稳定模态映射关系，严重降低变化检测的区分精度与鲁棒性。方法针对上述问题，本文提出一种融合树突神经元模型（dendritic neuron model，DNM）的跨模态重建无监督变化检测网络。该方法基于双边自编码器架构，构建多尺度卷积编码模块挖掘遥感图像多层级特征信息，并且创新性地引入仿生DNM建模复杂非对称的跨模态非线性重建关系，有效适配异质图像强模态差异。同时，设计动态区域掩码自适应约束机制，引导对抗损失与特征一致性损失约束稳定不变区域，规避变化像素对模态对齐学习的干扰，联合重建损失与结构相似性损失实现多约束协同优化。结果在Italy、Gloucester、California和Texas等四组公开的异质遥感图像数据集上进行测试，结果表明所提方法的总体检测精度分别达到0.973 4、0.957 1、0.924 5和0.989 0，能够有效平衡误检与漏检水平。消融实验进一步验证了动态区域掩码机制的有效性与必要性。结论所提算法能够在无人工标注条件下增强异质遥感图像的跨模态可比性，有效抑制“伪变化”和变化像素的学习干扰，提升无监督场景下异质遥感图像变化检测性能。本文已公开算法完整实现代码，开源地址为：https：//cstr.cn/10441.11.sciencedb.j00240.0009x。

Recent advancement of large language models (LLMs) has led to significant breakthroughs across various tasks, laying the foundation for the development of LLM-based speech translation systems. Existing methods primarily focus on aligning inputs and outputs across modalities while overlooking deeper semantic alignment within model representations. To address this limitation, we propose an Adaptive Inner Speech-Text Alignment (AI-STA) method to bridge the modality gap by explicitly aligning speech and text representations at selected layers within LLMs. To achieve this, we leverage the optimal transport (OT) theory to quantify fine-grained representation discrepancies between speech and text. Furthermore, we utilize the cross-modal retrieval technique to identify the layers that are best suited for alignment and perform joint training on these layers. Experimental results on speech translation (ST) tasks demonstrate that AI-STA significantly improves the translation performance of large speech-text models (LSMs), outperforming previous state-of-the-art approaches. Our findings highlight the importance of inner-layer speech-text alignment in LLMs and provide new insights into enhancing cross-modal learning.

Reliable manual annotations for Vietnamese logistics calls are scarce, and telephone channel degradation compounds errors in business-critical entities. Existing adaptation methods typically optimize aggregate transcription accuracy, use uniform token losses, or apply a fixed training schedule; few control parameter updates, entity-level supervision, and sample order together when labels are noisy. We address this problem by combining low-rank adaptation (LoRA), token-level entity-weighted cross-entropy, and validation feedback curriculum learning for Whisper-large-v3. A joint acoustic–semantic score ranks samples by signal quality, sequence length, entity density, and linguistic rarity. The curriculum boundary expands only after the validation loss improves. On 20,317 in-domain training utterances, the complete system achieves a word error rate (WER) of 0.0735 and a mean exact match entity accuracy of 92.1%. Compared with vanilla LoRA, it reduces WER by 4.92% and raises entity accuracy from 85.3% to 92.1%. Ablation results show that curriculum pacing and entity weighting each contribute to the final performance.